Understand GPT Image 2.5 before you write a prompt.
GPT Image 2.5 — also written gpt-image-2.5, ChatGPT Images 2.5, or OpenAI Image 2.5 — is OpenAI's image generation and editing model for ChatGPT and the API. It sharpens detail, renders more natural light and texture, keeps subjects faithful to reference photos, and follows editing instructions more reliably across a long conversation.
The two outer buttons leave this site. The official page belongs to OpenAI; the mirror site is an independent third-party service. Neither is operated by AI Tool Blog, and availability, quotas, and pricing can change, so confirm them on the page you land on.
See what GPT Image 2.5 actually outputs
Every image below comes from OpenAI's own announcement, so you can judge the rendering, style adherence, and layout handling before you spend time on prompts. Nothing here was re-generated or retouched by us.
Images are published by OpenAI and reproduced here to describe the model. Source: the official announcement
Change one thing, keep everything else
OpenAI's own before-and-after example shows the behaviour that matters most in production: the instruction changes the outfit, while the pose, framing, and background stay where they were.
One model family, two ways in
GPT Image 2.5 is the model behind ChatGPT's image generation and the gpt-image-2.5 models on the API. The ChatGPT name and the API name refer to the same generation, which is why you will see ChatGPT Images 2.5 and GPT-Image-2.5 used for the same release.
A generation and editing model
It handles text-to-image generation and image-to-image editing through the same family, so a prompt or a reference photo can start the work.
Two API variants
Flare trades precision for speed and is the default for most applications. Sunburst spends more time to reach higher accuracy for premium visual work.
Faster than the previous generation
OpenAI reports up to 50% lower image generation latency compared with Image 2.0, alongside sharper detail and more natural light and texture.
Built for iteration
Multi-turn editing follows instructions more reliably, so a long conversation keeps improving one image instead of degrading it.
Pick the variant that matches the job
Both variants were released in the API at the same time and expose the same interface, so switching between them is a model-name change rather than a rewrite.
| Variant | Where it focuses | Best for |
|---|---|---|
| GPT-Image-2.5 Flare | Quality, editing, and speed together, with up to 50% lower latency than Image 2.0. | Most applications: high-volume generation, interactive editing, and anything where waiting hurts the workflow. |
| GPT-Image-2.5 Sunburst | Higher precision at the cost of a longer generation time. | Premium visual workflows where detail and exactness matter more than turnaround. |
OpenAI describes Flare as the default choice for most applications. Neither variant is a separate product tier — they are two tunings of the same model.
GPT Image 2.5 versus the previous generation
These are the differences OpenAI calls out in the release notes, written as the short pairing that is easiest to check against your own results.
- Latency
- Up to 50% lower image generation latency than Image 2.0.
- Rendering
- Sharper detail, more natural light, and richer texture.
- References
- Subjects stay recognisably themselves when you move a reference photo into a new scene, style, or composition.
- Editing
- Instructions change the element you named and leave the surrounding subject, composition, and background alone.
- Multi-turn work
- Earlier edits hold up across a long conversation instead of drifting as the session grows.
- Layout and style
- Better handling of complex layouts, transparent backgrounds, and stated visual styles, including images that carry real-world information.
Output types covered by the official examples
The official announcement demonstrates the model across design, information, photography, and illustration work. These are the categories those examples fall into, each with a matching sample above.
Marketing and brand assets
Campaign visuals and branded material that follow a stated visual direction and keep their hierarchy.
Presentation and information graphics
Slides and diagrams that need to stay readable, with the structure you asked for rather than decorative padding.
Product and e-commerce imagery
Catalogue-style shots and cutouts, including transparent backgrounds for placing into an existing layout.
Series and repeated structures
Poster sets, stamp sheets, and sticker packs where one visual system has to survive across many items.
Photography-led scenes
Portraits, interiors, and location work where lighting and texture decide whether the result is usable.
Illustration and stylisation
Painterly, printed, and graphic styles where adherence to a reference style matters more than realism.
What OpenAI has and has not published
This page only states what the official release notes confirm. Where a number or limit is missing from that source, we leave it blank instead of filling it in from third-party resellers.
- Release date
- 8 September 2026
- API models
- GPT-Image-2.5 Flare (default, speed) and GPT-Image-2.5 Sunburst (precision)
- ChatGPT availability
- All ChatGPT tiers, ChatGPT Work, and Codex, on desktop, mobile, and web
- Latency
- Up to 50% lower than Image 2.0, per OpenAI
- Transparent background
- Supported, and named in the official release notes
- Content credentials
- C2PA metadata and an invisible watermark
- Price and per-tier quotas
- Not published on the announcement page
Two entry points, two audiences
The same generation is reachable from a chat interface or from code. Pick based on whether you need judgement in the loop or volume.
ChatGPT
Image 2.5 is available to all ChatGPT tiers, ChatGPT Work, and Codex users on desktop, mobile, and web. The interactive features — Sketch, templates, image comments, and prompt sharing — live here.
API
Developers reach the same generation through the Flare and Sunburst models, which is what production workflows, batch jobs, and product integrations build on.
GPT Image 2.5 FAQ
Are GPT Image 2.5 and ChatGPT Images 2.5 the same model?
Yes. ChatGPT Images 2.5 is the name used inside ChatGPT, and GPT-Image-2.5 is the name the API models use. They come from the same release, so a capability described for one applies to the other.
Should I use Flare or Sunburst?
Start with Flare. OpenAI positions it as the default for most applications because it combines quality and editing with up to 50% lower latency. Move to Sunburst when a longer generation time is worth more precision.
Is GPT Image 2.5 free in ChatGPT?
OpenAI states that Image 2.5 is available to all ChatGPT tiers, ChatGPT Work, and Codex users. It has not published per-tier image quotas, so the number of images you get depends on your plan and account and is best confirmed inside the app.
Do the images carry a watermark?
OpenAI says outputs continue to use C2PA metadata and an invisible watermark so generated images can be identified. The system card documents the evaluation method behind the safety work.
How much does the GPT Image 2.5 API cost?
The announcement page does not list figures; it links to OpenAI's API pricing page for current rates. Per-image numbers circulating on reseller sites are that reseller's own pricing, not OpenAI's.
Continue with a focused guide
This page is the overview. The two collections below take the same model into step-by-step tasks and into code, and every article links back here.
Learn GPT Image 2.5 task by task
Prompting, Sketch, templates, transparent backgrounds, character consistency, and comparisons against the models people ask about most.
Browse the free guides Developer docsIntegrate gpt-image-2.5 into your product
Model selection, request parameters, image editing calls, cost control, and the operational details that only matter once you ship.
Open the developer docsCapability, availability, and safety statements on this page come from OpenAI's official announcement for ChatGPT Images 2.5. Example images remain OpenAI's and are reproduced with credit and a link back. AI Tool Blog is independent and is not affiliated with OpenAI.
Latest GPT Image 2.5 articles
Browse every published GPT Image 2.5 article, from introductions to practical guides and developer documentation.
-
Designing UI Mockups and App Screens with GPT Image 2.5
Brief GPT Image 2.5 for a believable UI mockup: fix the frame, keep in-screen text short, hold one product across screens, and repair a single label.
Read article -
Sticker Packs and Transparent Emoji with GPT Image 2.5
Brief GPT Image 2.5 for a sticker pack: lock one character, build the set from a reference, ask for a transparent background, and fix one expression.
Read article -
GPT Image 2.5 vs DALL·E 3
GPT Image 2.5 and DALL·E 3 compared from OpenAI material only: what the retirement note covers, what the current generation documents, what stays unpublished.
Read article -
Making Infographics and Diagrams with GPT Image 2.5
Turn teaching and technical material into labelled diagrams with GPT Image 2.5: how to write the brief, set the frame, keep in-image text short, fix one label.
Read article -
GPT Image 2.5 Prompt Sharing
How prompt sharing works in GPT Image 2.5: what travels with a shared image, how others reuse the prompt with their own photos, and what is undocumented.
Read article -
GPT Image 2.5 vs Nano Banana 2
An official-documentation comparison of GPT Image 2.5 and Nano Banana 2: documented capabilities, access, quotas, and what neither vendor publishes.
Read article -
GPT Image 2.5 vs Seedream 5.0 Pro
GPT Image 2.5 and Seedream 5.0 Pro compared from official documentation only: documented capabilities, real trade-offs, and the numbers no vendor publishes.
Read article -
GPT Image 2.5 vs FLUX 2
GPT Image 2.5 and FLUX 2 compared from official documentation only: documented capabilities, open-weight trade-offs, and what neither vendor publishes.
Read article -
GPT Image 2.5 vs Ideogram
GPT Image 2.5 and Ideogram 4.0 compared from official documentation: typography, editable output, deployment, and what neither vendor publishes.
Read article -
Is GPT Image 2.5 Free
GPT Image 2.5 is included on all ChatGPT plan tiers, but no official page publishes an image quota. Here is what is confirmed and what is not.
Read article -
GPT Image 2.5 Sketch
Sketch lets you draw a layout inside ChatGPT and use it as the reference for the finished image. Here is the exact entry point and what the drawing controls.
Read article -
GPT Image 2.5 Templates
Templates in GPT Image 2.5 start you from a format such as Poster or Merch instead of a blank canvas. Here is the official entry path and its one limit.
Read article -
GPT Image 2.5 Comment Editing
Comment editing in GPT Image 2.5 pins an instruction to the exact spot it applies to. Here are the official steps for mobile and web, plus the stated limit.
Read article -
GPT Image 2.5 Character Consistency
GPT Image 2.5 keeps a character more recognisable across scenes, but OpenAI stops short of a guarantee. Here is what improved and how to work around the rest.
Read article -
GPT Image 2.5 Combine Images
How GPT Image 2.5 turns several input images into one frame: the documented input limit, how to label each reference, and why faces drift between rounds.
Read article -
GPT Image 2.5 Text Rendering
What OpenAI documents about text inside images, the four limitations it lists, and the official quoting and spelling rules that keep lettering legible.
Read article -
What Is GPT Image 2.5
GPT Image 2.5 is not one model ID but two: Flare and Sunburst. Here is how they differ, what the specifications allow, and which tasks each one fits.
Read article -
GPT Image 2.5 API Overview
An overview of OpenAI's GPT Image 2.5 image API, covering the Images API and Responses API, key parameters, authentication, and getting started.
Read article